Deep Principle · Scientific discovery LLM

Hydrogen-1.0

Hydrogen-1.0 is a scientific model from Deep Principle, built by post-training GLM-5.3 on chemistry, materials science and life sciences data, plus records and agent trajectories from the company's own laboratories. The company says it outperforms GPT-6-Astra on FrontierScience-Research, improves on GLM-5.3 on GPQA-Diamond and roughly matches Kimi-K3, though the text gives no scores for these. On internal tasks it beats GPT-6 Astra at maximum thinking by 5.13 points on scientific computing and 4.76 points on one-step forward reaction prediction, while Astra leads by 3.33 points on NMR structure elucidation. Demonstrations include writing and repairing a GROMACS polymer simulation workflow, growing an Ireland–Claisen reaction network of 27 structures through 30 reactions, and compiling a 12-vial Suzuki coupling protocol in a virtual lab environment that catches errors before execution. It also learns from electronic lab notebook records and explicit discovery trajectories. The text is cut off before any licence, weights location or release details, though the stated aim is an open-weight model.